14 ms·
AI bots are now better than humans at decoding CAPTCHAs
- vouaobrasil 3y agoI get the feeling that newer generations of CAPTCHAs will no longer be trying to filter out bots from entering human-made sites, but humans from entering bot-made sites.
- rgrieselhuber 3y agoTo be honest, it already feels like a resource denial effort. Reminds me of Paul Virilio’s description of systems that deliberately inhibit human speed.
- IggleSniggle 3y agoCAPTCHA is just usefully accidental punishment for evading surveillance capitalism while casually browsing the web. There are sites where getting through a CAPTCHA is literally impossible if you are using privacy controls, you just get stuck in an endless loop of CAPTCHA completion. But if you're "logged in" to the surveillance network, there's zero friction whatsoever.
- rgrieselhuber 3y agoIndeed.
- rvnx 3y agoFurthermore, we can easily imagine that in +/- 10 years, your device will ask you to "verify your identity" using your ID card and a selfie, and in return, you will get this juicy Web Environment Integrity token that acts like a key to unlock Captcha and pornographic websites.
- rightbyte 3y agoLike the ordering kiosks at McDonalds?
- EGreg 3y agoThat’s not hard to do There are always things computers do way better than humans. That’s why they have been created in the first place! Anyway, what will the new CAPTCHAs consist of? If the answer is “nothing”, then that’s the age of AGIs The CAPTCHA can ask humans to do a non-trivial research task, like look up the latest news and give an opinion, or actually appear on camera etc.
- joseda-hg 3y agoBoth of those are prone to be broken by a reasonably good AI, plus verification of either would be too computationally heavy to be cost effective (Also likely involving AI)
- brandonhorst 3y agoIt's nice that we named it so forward-thinking. Not a necessarily Completely Automated Public Turing test to Keep Computers Out.
- bandergirl 3y agoThat makes no sense. You just need an undocumented binary interface to keep most humans out, no need to create an HTML page and add a reverse-CAPTCHA.
- firtoz 3y agoBut the humans keep reverse engineering our APIs and making beautiful UX on top of them
- vouaobrasil 3y agoI meant that in the future, most sites will be made by bots and using them will be too hard for most humans...
- hinkley 3y agoHow good are bots at simulating human behavioral patterns these days? On my back burner I have a crowd-sourced data app and I keep wondering how I'm going to keep bots out. The ideas of shadow banning, throttling, or an approval queue for everything except known 'real' humans and new users that seem relatively human keeps popping up (eg, 2 approval queues for 'probably a bot' and 'probably not a bot')
- user764743 3y agoThis is very dependent on the bot and what it's running on.
- jsnell 3y agoDupe: https://news.ycombinator.com/item?id=37133485 https://news.ycombinator.com/item?id=37133485
- hilbert42 3y agoGood, they may become redundant and many of us would love that.
- londons_explore 3y agoThere are still plenty of good captcha approaches left, they're just more invasive. Things like "please type your credit card number here and pay a 1 cent fee". Or "send us a selfie with your passport". Or "Get 5 friends to vouch for you". Or "please log into a government website and oauth this app" Or "please log in with a gmail account that is at least 5 years old".
- JimtheCoder 3y agoAre we just making lists of things I'll never do?
- stjohnswarts 3y agoYeah talk about just finding another way to spend time after work :) . Guess I'd get more into my wood working or reading more books.
- _8j50 3y agoOr "turn on your webcam and say/do stuff" I have a feeling other humans verifying you mechanicalturk style one time is the future. But imho, the payment route makes sense and also this is the problem as random paywalled sites and web monetization. A cash equivalent bearer of token payment method is what is needed. 1) verify+lock funds 2) get bearer to pay funds or funds unlock in x minutes, cancelling the payment 3) token is presented to payment processor who will revoke the token, issue funds to payee with a new token matching that value paid 4) anyone holding that token can pay for stuff, you can just email or move tokens with usb drives or store them im a secure vault service I believe the main obstacle to solve so many web issues are KYC laws and lack of constitution amendment level laws that give the people right to trade using a bearer token (cash or not) and transfer funds peer-to-peer without a third party or disclosing their personal info (again, all like cash).
- mattnewton 3y ago> Or "turn on your webcam and say/do stuff" Probably <1 year away from beating automated detection of this reliably and 2-3 in fooling humans a high % of the time economically. I was at siggraph recently where I think I saw a whole room of papers of people’s approaches to doing just this problem of photorealistic faces saying arbitrary things.
- jordanreger 3y agoI'm not surprised. I input a 2FA time-based code with a 30 second refresh on it and by the time I got through the captchas, my code had expired. By ~1.5 minutes...
- indymike 3y agoThe approach that will work: 1. Have a service worth paying for. 2. Collect a payment with registration/account creation.
- mathgeek 3y agoCaptchas are often implemented on the payment forms themselves to prevent card testing.
- junaru 3y agoThat approach will not work at all. Majority shops have captchas since crawlers are like mini denail of service attacks on their resources and they continue day and night. You will not pay a single webshop to browse its inventory let alone dozens you currently check before buying anything. Like it or not this is what the internet runs on nowadays so killing that would have much bigger fallout than anything annoying captchas could do.
- sam0x17 3y agoOr really, proof-of-work based CAPTCHAs. I think inevitably that's the world we're eventually stuck in
- neilv 3y agoA bunch of news sites try that, but every link to them on HN (an Internet tech community you'd think would have a lot of people especially sympathetic to merit-based monetization of digital content) quickly gets a comment with a piracy link.
- fluoridation 3y agoMaybe the service news sites provide isn't worth paying for for the vast majority of incidental visitors.
- BeFlatXIII 3y ago…because no one who isn't already a subscriber is going to pay to read one article.
- super256 3y agoIt feels like captchas are mostly to stop humans from creating multiple accounts by hand nowadays.
- verteu 3y ago> reCAPTCHA: The accuracy [for humans] of image classification was 81% and 81.7% on the easy and hard settings respectively. Surprisingly, the difficulty appeared not to impact accuracy. I'm surprised, too. My solve rate (as a human) is <50% for reCAPTCHAs of the form "select all squares containing a motorcycle/bicycle". Does anyone have tips for solving them? (Is the rider part of the motorcycle? Should I select a mostly-empty square containing a single handlebar?) All I've determined is that the system prefers contiguous sets of ~6 squares.
- CalRobert 3y agoNot sure but it drives me crazy when I have to knowingly miscategorize something because it thinks a motorcycle is a bicycle and vice versa.
- bfeynman 3y agoCAPTCHAs are supposed to tell humans and robots apart, if robots are all doing better than humans, maybe we should flip the acceptance criteria to make sure you are not performing task at superhuman level, (until we train bots to do this). On an unrelated note, I have found captchas that don't even work and reprompt me all the time, I wonder if there is some naive filtering behavior they are applying.
- camel-cdr 3y agoI feel like they are sometimes already doing that, I had multiple CAPTCHAs where I thought I miss clicked, but got through anyways.
- microtonal 3y agoIt’s been like that for years, they probably use a subset for extra annotation.
- hunter2_ 3y agoI think only some parts of a CAPTCHA are challenging you (data is labeled sufficiently for mistakes to be considered a failed solve), and the other parts are still in the process of being labeled, which you are helping to accomplish, so those have no influence on the immediate outcome. At least that's how the old "type these two words" CAPTCHAs worked. It was crowdsourced human OCR of whatever text the machine OCR couldn't make sense of. I'm not sure if "find the bus/motorcycle/crosswalk/light" is the same way, but perhaps, and it does seem to offer leeway when there's only a few pixels of that item in the frame.
- jorvi 3y agoWhat’s sad is, I’ve started to predict how dumb other humans are at filling in CAPTCHA’s. Often there’s a piece of motorcycle, stair, traffic light, fire hydrant, or non-bus picture that users will mis-select, and I’ll do what I think the masses do. I used to choose correctly but being sent through 5 chains of CAPTCHAs is modern hell.
- 3y ago
- Workaccount2 3y agoThis is going to create a mess and make the internet probably suck a lot more. I don't have any solutions or ideas to float. Just painful acceptance that nice things are going to break.
- yreg 3y agoOne solution was the universally hated worldcoin. I hate it as well and don't think it could work, but the problem they meant to tackle is very real and we should talk about it. I think we are going to need some kind of certificates vouching for humans being humans.
- mewpmewp2 3y agoI wonder if there could be a service which you use with your real identity registered, but the certs it gives you are untracked to still allow for possibility of being anonymous. Although it seems like it should limit it to giving you only a few certs a month. But then bad actors could start buying those certs from people themselves. Still would make it much harder to spam requests.
- graypegg 3y agoI could totally imagine Verisign or something doing this. Imagine a world where you have to pay a yearly renewal fee to keep your humanity. They’d love it. Of course, it’s not a claim that the user is a human, but rather the user has a slightly difficult to obtain certificate that had to be requested by a human. Could slow spam down though.
- shkkmo 3y ago> a service which you use with your real identity registered, but the certs it gives you are untracked to still allow for possibility of being anonymous. The best you can do is pseudonymity. If the attestations are completely anonymous then you have no way of dealing with bad actors, like those who sell their digital identity to spammers.
- noAnswer 3y ago
- TheMagicHorsey 3y agoGood. Nobody likes captchas.
- ethanbond 3y agoI like websites not filled top-to-bottom with spam though.
- mensetmanusman 3y agoWebsites should just ask us how we feel about flipped turtles.
- dreadlordbone 3y agoBut you're not helping. Why is that, Leon?
- hackingonempty 3y agoThis is why techies will not be able to defeat WEI. It will be the replacement for CAPTCHA. Every place using a CAPTCHA now will switch to requiring cryptographic attestation that you're running a clean copy of FAANG stack on approved hardware. The only effective non-human bots will be physical robots using the analog hole.
- firtoz 3y agoWhat is WEI?
- cute_boi 3y agoDRM of Web . https://arstechnica.com/gadgets/2023/07/googles-web-integrity-api-sounds-like-drm-for-the-web/ https://arstechnica.com/gadgets/2023/07/googles-web-integrit...
- doylio 3y agoThere is a knee-jerk aversion to the word blockchain on hackernews, but crypto folks have been thinking a lot about decentralized sybil resistance. If the choice comes down to WEI or a blockchain based solution, how would you feel about that option?
- GhostWhisperer 3y agowhat does a blockchain bring into the discussion that you think parent would be okay with? if the choice comes down to two options, what makes you think it would be between wei and blockchain i think there might be some confusion here simply because parent used the work cryptographic, but this has nothing to do with blockchain
- chrismarlow9 3y agoIt will get bypassed easily. The lengths spammers will go to for "organic" advertising that is automated doesn't really have limits. Fine, add WEI, spammers will write code that literally issues native events to a real browser on a farm of devices. I actually saw this discussion in the WEI proposal GitHub regarding Chrome extensions. What gets a pass? Captcha was never really the limiting factor for spammers. It was generating human readable content at scale (comments, reviews, responses, etc). But with AI that's not an issue anymore.
- jokethrowaway 3y agoGreat, can't wait for an AI powered captcha solving, ad blocker, EU cookies popup closer application. Maybe we'll get back the good old web
- dkersten 3y agoAnd good old web (well, that intermediary persons were many things didn’t yet use captchas but bots were already prevalent ) levels of spam and flooding of comment sections with links to advertisement and scams. Sounds great. I hate captchas as much as the next person and have long suspected that bots can solve them better than I can, but I hate the comment sections and forums made useless by spam messages even more.
- wouldbecouldbe 3y agoAs far as I understand at least for Google's captcha. They mostly checks check mouse browser details, activity & reaction time to check if your human. The challenge itself is only a very last resort. Didn't they use that for training data at some point?
- siva7 3y agoThe whole premise of captchas falls apart in the AI age and i’m glad not having to see anymore google captchas & co in the near future
- kmeisthax 3y agoCompletely Automated Public Turing test to train Computers to imitate Humans Absolutely
- Aerbil313 3y agoCaptchas were doomed since the transformer AIs are invented anyway.
- wintermutestwin 3y agoThe only thing it seems that CAPTCHAs are good for these days is to discriminate against VPN users. There are several sites (including govt) that put me in an endless CAPTCHA loop.
- GaggiX 3y agoI fortunately never had a problem using Mullvad VPN, only when using Tor.
- wintermutestwin 3y agoOn Mulvad as well and I have run into quite a few unusable sites.
- GaggiX 3y agoI do not really have this experience and I wonder why, even when I use the Mullvad browser I do not have any problems. Can you give me an example of a site that is not usable?
- GhostWhisperer 3y agocould it be you're leaking identifiable bits through the browser?
- GaggiX 3y agoAre you talking about canvas fingerprinting and stuff like that?
- GhostWhisperer 3y agoyes https://www.amiunique.org https://www.amiunique.org https://coveryourtracks.eff.org https://coveryourtracks.eff.org
- ChatGTP 3y ago
- ChrisArchitect 3y ago[dupe] Lots of discussion here 2 days ago (and that wasn't even the bunch of posts from a week ago) https://news.ycombinator.com/item?id=37133485 https://news.ycombinator.com/item?id=37133485
- ryandvm 3y agoSimple solution: The humans are the ones failing the CAPTCHA
- atentaten 3y agoThis can be used to support the idea of requiring digital identification to access the internet.
- klyrs 3y agoI have stopped trying to get perfect scores on captchas. Instead, I've been testing to see how many errors I can get away with. This is a frivolous act of resistance, as the bots clearly don't need my help anymore.
- yieldcrv 3y agoSame, I can mentally visually a heat map of what other humans would have clicked on the most and just do the most central elements
- Vicinity9635 3y agoYup. I deliberately half-ass it. Seems to work even better and faster than trying to get it perfectly correct.
- alaxapta7 3y ago> faster It's purely anecdotical, but I'm convinced that Google CAPTCHA (which I don't see more often only because I rarely use Google Search) punishes me for being fast. I can half-ass it or do a perfect score, but if I'm done with it in less than 3 - 5 seconds, I'll only get another puzzle as a reward.
- stjohnswarts 3y agoto me it seems like there is a certain area size of motorcycle/bicycle/traffic light and below that it doesn't matter if you pick the block or not.
- niyaven 3y agoWhich is why I'm always using buster[1] when facing captchas. https://github.com/dessant/buster/ https://github.com/dessant/buster/
- ornornor 3y agoThe captcha buster extension[^0] along with the service on you computer to move the mouse for you works very very well. It uses google’s TTS (afaik) to transcribe the audio captcha. It’s google verifying another google service works well. I find it very satisfying to not provide my labor to train googles computer vision corpus but instead have the snake bite it’s own tail. Anyway, I highly recommend buster, I barely notice captchas anymore with it. [^0]: https://github.com/dessant/buster https://github.com/dessant/buster
- bradly 3y agoKind of funny as OpenAI's CAPTCHA involved me picking a person's seat on a plane. I wonder if they test it against their models.
- modzu 3y agoalways have been
- squokko 3y agoI remember in the text captcha days there came a point where I just couldn't do them anymore. The letters were so distorted it would take me 10+ tries.
- falcor84 3y agoHow about "Assess whether this LLM output is a hallucination?"
- yieldcrv 3y agoWhen people tell me unsubstantiated things to uphold their beliefs, I’ve started to point out “if an AI said that, we would say it was hallucinating” and honestly, I cant find a reason to privilege human bullshitting over AI’s
- hidelooktropic 3y agoThe very paradigm of gating by cognitive traits is inherently a fragile one when we are actively working to bring computers in or beyond parity with our own. We will probably rest back on the old ways of doing things where we use authentication and trust-building with verified accounts to throttle and gate what users get what privileges. The evolving coherence of stronger and less siloed identity attestation seems like an obvious way to go here, at least looking way back from the past where we are now.
- jdlyga 3y agoThe goal of modern CAPTCHAs is to determine if the user is "acting like a human". Deciphering warbled, blurry text worked in the days when OCR was difficult. As we can see, that rule no longer applies. Now, it should be about making sure behavior is typical or atypical. Is this a human browsing the web, or is this a script using AI to pass captchas?
- akeck 3y agoSomething I've wondered about is how CAPTCHAs are ADA-compliant. Ironically, this tech may end up helping people who are impaired from doing CAPTCHAs use websites. Just the other day, I watched a friend of mine who has vision health issues do several CAPTCHA rounds to use a state website. Edit: There's various efforts: https://captcha.com/captcha-accessibility.html https://captcha.com/captcha-accessibility.html
- jakubmazanec 3y agoAs I asked in similar thread the other day, what about proof of work based CAPTCHA like https://github.com/mCaptcha/mCaptcha https://github.com/mCaptcha/mCaptcha - which I didn't actually see on any site. Is it used? Since CAPTCHAs can be solved by bots, at least make it more costly for them.
- itissid 3y agoCompanies that want Physical Humans to watch what is on their websites(ads+content) and wall off said content at the same time. Every one of these companies is on borrowed time. Eventually AI will be able to read frame buffer and simulate mouse movements to a tee and get what any one wants, Tor networks and VPN networks will be used to circumvent API restrictions and everything else that blocks people will be learnt by the AI. The only factor will be cost of compute that will host the AI and how much data you want. If there was no profit motive(ads) or its correlates(attention), none of this would be necessary. Companies would just provide utility to make people's lives easier. The most utilitarian company would win. The only reason big tech needs to make this much money is share holder "value"(code for: 'we need to make our execs and engg rich so that other companies don't grab them'). This is a massive snowball; Its crash-only thinking(google the term).
- ekanes 3y agoSort of. Ads as a business model exists because many people don't want to pay companies directly (through subscriptions etc). For the company to exist, they have to make money, so they sell people's attention. Unfortunately, we simply haven't found a better way yet.
- jfengel 3y agoI'm no fan of capitalism or the ham-fisted way companies try to control their intellectual "property", but "shareholder value" is what gets content created under the system we've got. Somebody ponied up money for a content creator to sit down and do the work. They wouldn't do that if they weren't hoping for a return on that investment. Me, I'd love to see a world where everybody got enough money to live on, and then got to sit down and create whatever content they want, for the heck of it. They don't own it; we all pay for it and we all get to enjoy it. But that's not happening any time soon. So I think we're stuck with the problem of intellectual "property" being a thing, and companies trying to artificially limit access to it despite knowing that there are a million ways around it. They're just going to hope that most people, most of the time, would rather take the legal and official route, if it's not too burdensome.
- AtNightWeCode 3y agoDelete all captchas everywhere. While at it. Delete all challenges. I am tired of wasting time into all this bs. If we did put the same amount of time into real solutions like rate-limiting and actually making endpoints secure there would be no problem to begin with.
- znpy 3y agoI’m failing captchas almost 100% of the time lately. Particularly those where you have to slide the part of the image. I guess I’m not human anymore?
- shric 3y ago> The bots’ accuracy ranges from 85-100%, with the majority above 96%. This substantially exceeds the human accuracy range we observed (50-85%),” the research paper read. Wait, did they use humans or computers to measure the accuracy of the solutions?
- squarefoot 3y ago"Please, fail this CAPTCHA to give proof that you are human"